Phoenix Camelback Corridor AI SEO, AEO and GEO

Meet the robots that read a firm's site for AI tools

Each AI tool sends its own crawler, and a robots.txt rule written for one can quietly shut out another, so start by knowing who's knocking.

Robots files near Camelback Road, like everywhere else, collect junk over the years. A developer blocks one bot during a launch, a plugin adds another rule, and three redesigns later nobody on staff knows what the file says. It's one of the first things we open on any site, because a single stray line can undo months of careful writing.

The fix is rarely hard. It's a text file with a handful of lines. The hard part is deciding on purpose which tools you want reading your pages, then making sure one named person owns that file from here on. Everything else in AI search work sits on top of that one small choice, so we like to settle it before anything else.

A quick note on scale. A small firm's site might get a visit from each of these crawlers every so often, and none of them will slow your pages down in any way you'd notice. So server load barely enters into it. The real question is which tools you want describing your firm to the next person who asks.

A coffee-break robots.txt check you can run today

Checking what your site allows needs no developer, and it takes about as long as a cup of coffee at your desk.

1. Open your robots file Your robots file lives at a fixed spot: your domain, a slash, then robots.txt. Paste that into a browser. A page-not-found error means you don't have one, so nothing is blocked at that level. Any other error is worth a quick email to your host.
2. Look for a blanket block Scan for a Disallow line with a lone slash under User-agent with a star. That line shuts every polite crawler out of the whole site, and it sometimes survives from a staging copy.
3. Search for the AI crawler names Press Ctrl and F and look for GPTBot, OAI-SearchBot, PerplexityBot and Google-Extended. Jot down which ones the file allows, blocks or never mentions.
4. Ask your host about the firewall Robots rules aren't the only gate. Google's AI features guidance lists crawler access at the CDN and hosting level right beside robots.txt, so ask whether bot protection turns crawlers away.
5. Inspect one page in Search Console Run your About page through the URL Inspection tool. It shows whether Google has the page indexed and when Googlebot last crawled it.
6. Write the choice down and give it an owner Decide which tools you want reading your site, put that in one sentence and name the person who approves changes to the file. Surprise edits are how firms vanish without noticing.

If you only do one step, do the firewall question. Hosting companies and security plugins have grown more aggressive about bots, and some block AI crawlers by default with a single checkbox. The robots file can say welcome while the firewall slams the door, and nothing on your end looks broken. One short email to your host gets you the answer.

Once you've checked, write the date at the top of your notes. When we audit a site, we like to see when the file last changed and who changed it. A robots file with no history is how a well-meaning intern blocks the whole site on a Friday and nobody notices until the phones go quiet.

When a firm's words only appear after the scripts run

Some AI crawlers read only the raw HTML your server sends first, so text that shows up only after JavaScript runs may never reach them.

Plenty of modern sites build the page in the browser. The server sends a thin shell, then scripts pour in the headings, the services list and the team bios. People never notice, because their browser runs the scripts in a blink. A crawler that skips scripts sees the shell and moves on, like a guest who leaves before dinner is served.

Here's the quick test. Open a page, then open its source instead of the page itself. Search the source for a phrase you can see on screen, like a partner's name or your main service. Found it in the source? Then even a script-blind crawler can read it. If it's missing, that fact lives only in JavaScript, and it's worth a chat with your developer.

• Your firm's legal name and the name people actually call you
• Each service you offer, in the words a client would type
• Principal names and current titles
• The office area, plus whether clients visit in person
• A phone number and a way to book
• Anything you'd be upset to see an AI tool get wrong

That list is what we check for in the raw source first. If all of it is there as plain text, you're in decent shape no matter which crawler shows up. If some of it only appears after scripts run, we'd move those facts into the server-rendered page before touching anything fancier. It's rarely a rebuild. Often it's one template.

When a name partner leaves, the old firm name keeps talking

Professional firms rename more often than most businesses, and AI tools can keep repeating the old name long after the sign changes unless you leave a clear trail.

The formerly-known-as line is the hero here. A client who still has the old name on a sticky note searches for it and finds the new firm. A crawler reading the About page sees one sentence tying both names together. Without that sentence, a tool has two firms that look alike and no reason to connect them, so it may keep describing the one that no longer exists.

Firms often resist that line because the split was awkward. We get it. Skip the drama and the backstory. The old name, the new name and the year are enough. Clients mostly want to know they've found the right place, and a tidy trail tells them so before they ever pick up the phone.

We'd also look at email signatures, PDFs of old engagement letters on the site and the footer of every page. Old names hide in odd corners. Each one a crawler finds is another vote for the version of the firm that no longer exists, so we sweep them up in one pass.

• Keep the old domain renewed and pointing page by page to the new site.
• Add one formerly-known-as sentence to the About page, with the year.
• Point departed partners' pages to the current team page instead of an error.
• Update the Google profile, LinkedIn and your key directories in the same week.
• Ask your bar association, state board or trade group to update its listing.
• Add the old name to your monthly prompt check and watch it fade.

Partner bios a summary can't easily misread

In a professional firm the people are much of what clients buy, so each bio needs dated, checkable facts a tool can repeat without inventing a title.

• Full name as it appears on the person's license or registration
• Current title at the firm, and the year they joined
• What they help clients with, in plain client words
• What they don't handle, so nobody books the wrong person
• Any license a client can look up, plus where to look it up
• A photo caption with their name, since images get read too

Titles cause more trouble than you'd think. One page says senior partner, LinkedIn says managing partner and an old directory says associate. A summary picks one, and it may pick the stale one. We line up every bio against the person's own LinkedIn and the firm's letterhead, then fix the site first, since it's the version you control completely.

Two people with similar names are another trap. Maybe your partner shares a name with someone famous, or with a professional in another state. The bio should then carry details that set them apart, like their field and the office they work from. It feels fussy. It also keeps a tool from blending two careers into one confident paragraph about the wrong person.

We also look at who wrote the bio. Many were drafted years ago by someone who's since left, and they drift out of date one promotion at a time. We'd rather give each partner a five-minute read of their own page every quarter than let a summary repeat a job title they stopped using two years ago.

An About paragraph an answer engine can quote

AI tools summarize firms from the sentences they find, so an About paragraph full of fog gives them nothing solid to repeat about you.

Read the before column out loud and try to answer a simple question: what does this firm do? You can't, and neither can a summary. Every line could belong to any firm in any city. So a tool either skips it or fills the gap with something it found elsewhere, like an old directory entry.

The after column is plain to the point of boring, and that's the charm. Each sentence stands on its own, so a tool can lift any one of them without the others. It also says what the firm doesn't do. That line saves a partner from polite calls about audits, and it stops an AI answer from guessing.

We usually write this paragraph with a partner on a call rather than from a questionnaire. Ask someone what the firm does and they'll often say it perfectly in conversation, then freeze when they have to type it. We capture the spoken version, tidy it and send it back for a yes or a tweak.

Organization markup, minus the mystery

Organization structured data restates your firm's basic facts in a tidy code format, and Google doesn't make any single property mandatory.

• name: the exact name on your door and your letterhead
• url: the home page address people should land on
• logo: a clean image file of your mark
• address and telephone: the same ones printed on your Contact page
• sameAs: links to profiles that describe the same firm, like LinkedIn
• description: two plain sentences that match your About paragraph

One more thing to keep straight. Google says no special schema is needed to appear in its AI features, so this markup won't get you in on its own. We add it because it restates your facts in a format machines read easily, and because it's cheap to keep accurate. When a partner leaves or the phone number changes, the markup gets updated the same afternoon as the page.

If your site runs on a common platform, a plugin may already be writing some of this markup for you, and sometimes it's writing the wrong thing. We check what's there before adding anything, because two conflicting blocks of markup are worse than none. Google's free Rich Results Test is a handy way for anyone to peek.

Running a prompt test that actually tells you something

One prompt on one morning proves very little, so we run the same small set of prompts the same way each month and write down every answer.

We run the set in each tool the firm cares about, signed out where we can, with the same wording every time. For each answer we record the date, the tool and mode, what it said about the firm and which sources it cited. The cited sources are the gold. They tell us exactly which page or listing is feeding a wrong answer.

Don't expect every month to look better than the last. Tools change how they answer, and a good month can be followed by an odd one. What we look for is direction over a season: fewer wrong facts, more citations of your own site and fewer mentions of that dusty directory page.

We also keep the prompt set short on purpose. Five prompts run carefully every month tell us more than fifty run once. When a firm adds a new service or loses a partner, we add one prompt about that change and retire one that's been steady for a while.

AEO, GEO and llms.txt, answered without the jargon

The acronyms are new, but the questions behind them are old ones about being found and described accurately, so here are straight answers.

Is AEO or GEO really different from SEO?

Mostly it's the same work with a sharper focus. Google says the usual SEO basics still apply to its AI features. The difference is emphasis. You care more about whether each fact about your firm is stated plainly, matches everywhere it appears and sits somewhere a crawler can read without running scripts.

Should we publish an llms.txt file?

For Google, it won't matter. Google says it doesn't use llms.txt or similar files for Search, including its AI features, and that keeping one for other systems neither helps nor hurts you there. If another tool you care about asks for one, it's cheap to add. We just wouldn't build the plan around it.

Should we block GPTBot?

That's a business call about model training, and sensible firms land on both sides. OpenAI treats it separately from OAI-SearchBot, so blocking training doesn't have to mean vanishing from ChatGPT search. Pick a side on purpose, write it down and make sure your robots file says what you meant.

How often should we rerun the prompt test?

Monthly is plenty for most firms. Weekly checks mostly catch noise and burn staff time. Run it sooner after a big change, like a partner leaving or a service being dropped, so you can see how long the old fact hangs around. Keep the wording identical each time so months compare fairly.

Will a few ChatGPT visits really matter to a professional firm?

Probably not in volume. For most firms it's a trickle. But each of those visitors already heard your name from an AI tool before arriving, and many are close to calling. We'd watch which page they land on and whether they get in touch, and judge the channel on that.

Here's the kind of slip the monthly prompt test catches. Say the hypothetical accounting practice stopped doing audits a few years back, but an old services page still lists them. Ask a chatbot whether the firm handles audits, and it may cheerfully say yes. The repair has nothing to do with acronyms. You delete the stale line and add one plain sentence saying the firm doesn't do audits. Then you rerun that prompt next month and see whether the answer has caught up.

Who handles what between your team and ours

AI search work moves fastest when your team owns the facts and we handle the plumbing, so here's how we'd divide it with a Camelback firm.

Your side confirms the facts: services, people, titles, visit details and anything you've stopped doing. Only you can say those are true, and a principal's sign-off is what makes an answer worth publishing. Our side handles the robots file, the raw-HTML check, the markup, the About rewrite and the monthly prompt log.

We're an Indianapolis agency, so a Phoenix engagement would run over calls and shared files. There's no Phoenix firm on our client list today, so we'd show you each finding with its receipts: the robots lines we found and the exact prompts we ran. That way you can judge the work yourself.

In practice, your part takes a principal about an hour up front and a few minutes a month after that. Our part runs quietly in the background. When we sit down each month, we go over what the tools said about you and what we changed. Then we agree what to look at next, in plain words and without a wall of charts.